quantifying the impacts on biodiversity of policies for carbon sequestration in forests

Ecological Economics 40 (2002) 71–87

ANALYSIS

Quantifying the impacts on biodiversity of policies forcarbon sequestration in forests

Stephen Matthews a, Raymond O’Connor a, Andrew J. Plantinga b,*a Department of Wildlife Ecology, Uni�ersity of Maine, Orono, ME 04469, USA

b Department of Agricultural and Resource Economics, Oregon State Uni�ersity, Cor�allis, OR 97331, USA

Received 27 April 2001; received in revised form 16 October 2001; accepted 19 October 2001

Abstract

There is currently a great deal of interest in the use of afforestation (conversion of non-forest land to forest) toreduce atmospheric concentrations of carbon dioxide. To date, economic analyses have focused on the costs of forestcarbon sequestration policies related to foregone profits from agricultural production. No studies have examinedadditional costs or benefits associated with impacts on biodiversity. The main objective of this paper is to estimatethe changes in farmland and forest bird populations that are likely to occur under an afforestation policy.Econometric models of land use are used to simulate the response of private landowners to subsidies for tree plantingon agricultural land. We evaluate subsidies that achieve conversion of 10% of the total agricultural land in each ofthree U.S. states (South Carolina, Maine, and southern Wisconsin). Bird density estimates are derived for 615 specieswith data from the national Breeding Bird Survey. Percentage changes in agricultural and forest land for each countyare applied to county-level estimates of bird densities for farmland and forest birds. Despite considerable spatialvariation in agricultural land conversion rates and farmland bird distributions within these states, statewide losses offarmland birds were relatively uniform at 10.8–12.2%. Increases in forest bird populations, however, variedsubstantially between states: 0.3% in Maine, 2.5% in South Carolina, and 21.8% in southern Wisconsin. Surprisingly,a net loss in total bird populations results in all three states (−2.0% in Maine, −2.3% in South Carolina, and−1.1% in southern Wisconsin), despite the prevailing wisdom as to bird-rich forests. The loss is due to thecoincidence of centers of high farmland bird richness and low forest bird richness with areas economically suited toconversion. Additional gains in forest species may result, however, if afforestation within the economically optimalcounties is concentrated to fill in existing forest fragments presently suffering avian losses to edge predators. Ourresults thus show that assessments of the biological consequences of afforestation for carbon sequestration mustconsider both current land cover and the distributional patterns of organisms as well as the policy’s conversion goal.© 2002 Elsevier Science B.V. All rights reserved.

Keywords: Carbon sequestration; Avian abundance; Econometric models; Wildlife models; Land-use change

www.elsevier.com/locate/ecolecon

* Corresponding author. Tel.: +1-541-737-1423; fax: +1-541-737-1441.E-mail address: [email protected] (A.J. Plantinga).

0921-8009/02/$ - see front matter © 2002 Elsevier Science B.V. All rights reserved.

PII: S 0921 -8009 (01 )00269 -5

mailto:[email protected]

S. Matthews et al. / Ecological Economics 40 (2002) 71–8772

1. Introduction

The Kyoto Protocol to the Framework Con-vention on Climate Change, adopted by a major-ity of the world’s nations in December, 1997, setsspecific targets and timetables for the reduction ofgreenhouse gas emissions by Annex I (industrial-ized) countries. There is currently a great deal ofinterest in converting non-forest to forest land(afforestation) to offset carbon dioxide (CO2)emissions. Trees and other forest vegetation pho-tosynthesize CO2 to yield carbon, and sinceforests generally store more carbon than land inother uses (e.g. agriculture), afforestation canachieve a reduction in net greenhouse gas emis-sions. Article 3.3 of the Protocol states that car-bon sequestered as the result of human-inducedafforestation, reforestation, and deforestation isto be included in the emissions inventory used todetermine a nation’s compliance with its treatyobligations.

The decision to pursue an afforestation strategydepends, in part, on the costs of afforestationrelative to costs of alternative approaches such asimproving energy efficiency, switching to cleanerfuels, as well as other methods of carbon seques-tration (National Academy of Sciences, 1992;Holdren and Lee, 1999). A number of authorshave estimated the marginal costs of sequesteringcarbon in forests.1 For example, Plantinga et al.(1999) estimate econometric models of land use inwhich the shares of land allocated to forestry andagriculture are functions of net returns to alterna-tive uses and other decision variables. The fittedmodels are then used in a simulation of a subsidyprogram for afforestation. The subsidies increasethe relative net returns to forestry, which increasesthe area of land allocated to forest and theamount of carbon sequestered. Marginal costschedules are constructed by arraying subsidiesper unit of carbon against total carbonsequestered.

In general, previous studies find that the costsof carbon sequestration in forests are comparableto, and in some cases lower than, costs of alterna-tive mitigation and abatement approaches. How-ever, these analyses are focused solely on theopportunity costs of agricultural production. Animportant issue not considered in these studies isthe impact of the resulting land use changes onbiodiversity.2 Although agricultural land is gener-ally regarded as purely an anthropogenic habitat,it is, in fact, a significant resource for a variety ofspecies of conservation interest (e.g. for grasslandbirds) (Herkert, 1994; Vickery et al., 1994). Simi-larly, any advantages in the form of enhancedpopulations of forest species that might resultfrom afforestation are of relevance to conserva-tion efforts, particularly in the case of neotropicalmigrant birds, many species of which aremarkedly declining in numbers (Robbins et al.,1989b; Robinson et al., 1995).

Thus, a more comprehensive analysis of carbonsequestration costs would consider not only fore-gone profits from agriculture, but the additionalenvironmental benefits and costs associated withafforestation. In the present paper, we estimatethe changes in bird populations likely to ariseunder the carbon sequestration policy modeled inPlantinga et al. (1999). Our specific objective is todetermine the percentage changes in farmland andforest birds resulting from a policy that achievesconversion of 10% of the total agricultural land ineach of three U.S. states (South Carolina, Maine,and southern Wisconsin). We use birds as a tem-plate for other biodiversity calculations in thiscontext because the taxon is so data-rich, but themethods developed here can be extended to othertaxa, although with less reliable data. Given cur-rent momentum toward the use of carbon man-agement strategies to address global climatechange, we assume that carbon sequestration isthe primary policy objective. However, our studydevelops the tools needed for analysis of policies

1 Among the studies providing marginal cost estimates areMoulton and Richards (1990), Adams et al. (1993), Richardset al. (1993), Parks and Hardie (1995), Adams et al. (1999),Alig et al. (1997), Plantinga et al. (1999), Stavins (1999),Newell and Stavins (2000), Plantinga and Mauldin (2001).

2 Afforestation of agricultural land may have other environ-mental impacts. For example, in regions where intensive agri-culture is practiced, afforestation typically reduces soil erosionand the contamination of ground and surface water by agricul-tural chemicals.

S. Matthews et al. / Ecological Economics 40 (2002) 71–87 73

with multiple objectives. This is a first step to-wards the ultimate development of a nationalcarbon sequestration strategy designed to mitigateclimate change as well as to achieve other nationalenvironmental goals.

2. Materials and methods

2.1. Land-use change in response to subsidies forcarbon sequestration

In an earlier study, Plantinga et al. (1999) simu-late the response of private landowners to subsi-dies for carbon sequestration in forests. In thepresent study, we analyze the biodiversity impactsof the land-use changes associated with the af-forestation policies. We provide a summary of themethods used in the Plantinga et al. (1999) studyand present the results relevant to the currentanalysis. Readers are referred to the original studyfor more details.

Plantinga et al. (1999) simulate carbon seques-tration programs in Maine, South Carolina, andWisconsin. These states were selected becausethey represent a broad range of current land-usepatterns, physiographic conditions, and apparentopportunities for afforestation. Maine is a heavilyforested state with little additional land availablefor conversion to forest. In contrast, South Caro-lina and Wisconsin have large amounts of agricul-tural land that potentially can be afforested.Maine and Wisconsin are northern states withshort growing seasons relative to South Carolina.Southern pine tree species, valuable for lumberand plywood production, are abundant in SouthCarolina. Maine and Wisconsin have a mix ofhardwood species (e.g. oak, maple, birch) andsoftwood species (e.g. spruce, fir) used in paperproduction.

Econometric land-use models were estimatedusing standard methods developed in Lichtenberg(1989), Wu and Segerson (1995), Hardie andParks (1997). The county shares of land in privateforest (s it

f ), agricultural uses (s ita), and urban and

other uses (s itu) are specified as logistic functions of

exogenous variables (Xit):

s itf =

e��fXit

1+e��fXit+e��aXit, s it

a =e��aXit

1+e��fXit+e��aXit,

s itu =

11+e��fXit+e��aXit

, (1)

where i indexes counties, t indexes time, and �f

and �a are vectors of parameters to be estimated.The three land-use shares account for all land inthe county, implying s it

f +s ita +s it

u =1 and thatone of the shares is redundant. The additivityconstraint is incorporated into Eq. (1) by express-ing s it

u in terms of the remaining shares (i.e. s itu =

1−s itf −s it

a). The exogenous variables include thecounty average per-acre net return to forestry; thecounty average per-acre net return to agriculture;county population density, which controls for thediversion of land to urban and other uses; com-posite land quality measures, including the aver-age quality of land in the county and theproportion of the county’s land in the highestland quality classes; and a constant term and timedummies.

Separate models were estimated for each stateusing pooled time-series and cross-sectional data.Data were collected for all 16 counties in Mainefor the years 1971, 1982, and 1995, all 46 countiesin South Carolina for the years 1986 and 1993,and 49 counties in the southern two-thirds ofWisconsin for the years 1983 and 1996. Only thesouthern counties of Wisconsin were included be-cause much of the land in northern Wisconsin ispublicly-owned and already forested. See Ref.Plantinga et al. (1999) for details on the econo-metric procedures used to estimate Eq. (1) and theestimation results.

The land-use models were then used in a simu-lation of carbon sequestration programs. The ba-sic approach was to simulate per-acre subsidies toforestry by augmenting the corresponding net re-turn measure in the econometric model. This im-plied increases in forest area and declines inagricultural area relative to land use in the base-line. Simulations were conducted for different lev-els of a per-acre subsidy and the correspondingland-use changes were converted to carbon unitsusing yield functions developed by Birdsey (1992).A marginal cost schedule was constructed by ar-raying the subsidies—expressed in dollars per


unit of carbon—against total carbon sequestered.The highest marginal costs were estimated forMaine, followed by South Carolina and, lastly,Wisconsin. The low costs in Wisconsin were dueto the relative abundance of marginal lands withlow opportunity costs for agricultural production.Opportunity costs were sufficiently low in Wis-consin to more than offset the somewhat highercarbon sequestration rates in South Carolina.

In the current study, we focus on land-usechanges under scenario 1 in Plantinga et al.(1999). This scenario runs for 60 years beginningin 2000. In the baseline, all of the exogenousvariables in the econometric model (net returns,population, etc.) are held constant at mid-1990svalues and no timber harvesting is permitted onland enrolled in the program. Only agriculturallands are eligible and land must remain in theprogram for 10 years. In exchange, participatinglandowners receive a per-acre payment plus thecosts of tree establishment. Since all the exoge-nous variables are constant in the baseline andsubsidy levels remain constant over time, landenrolled in the 1st year of the program remainsenrolled for the duration of the program. In sce-nario 1, the subsidy is uniformly applied acrosscounties. Accordingly, marginal enrollment costsare equated across counties and the total cost ofachieving a given amount of land conversion isminimized.3

2.2. Bird data

Maine, South Carolina, and Wisconsin providediverse settings in which to study impacts onbirds. Bird populations differ substantially be-tween the three states, with South Carolina hav-ing a large component of year-round residentspecies while the avifauna of Wisconsin andMaine have a much higher proportion of migrant

species. Bird populations in the three states differmarkedly as to the environmental and land covervariables associated with their prevailing levels ofspecies richness (O’Connor et al., 1996), providingan ecological diversity paralleling the economicdiversity described above.

The bird data for the three states were derivedfrom the national breeding bird survey (BBS), abird population monitoring program conductedannually since 1966 in the United States andCanada, currently by the Biological ResourcesDivision of the U.S. Geological Survey and by theCanadian Wildlife Service. The scheme is adminis-tered by USGS staff at the Patuxent WildlifeResearch Center in Laurel, Maryland, and cur-rently acquires bird data from some 4000 routesacross the continent, though not all routes aresurveyed annually (Robbins et al., 1989a). Thesurvey focuses on diurnal birds that can becounted along a pre-determined route on sec-ondary roads (3 min counts of all birds detectedat 50 stops along a 25 mile route). Crepuscularand nocturnal species, and species restricted tooff-road habitats, are, therefore, not surveyed.The survey was designed to obtain representativeresults across North America, within the con-straints of this survey protocol, and a recent peerreview concluded that the scheme results in datathat, with only minor biases, largely meet its goals(O’Connor et al., 2000). O’Connor et al. (1996)extracted a set of 1200 representative BBS routesthat had frequent and high quality surveys overthe period 1981–90, and determined for eachroute the incidence of each species (the proportionof surveys along the route that had recorded thespecies). Yang et al. (1995) interpolated theseincidence data for each of the 615 individualspecies to obtain an abundance surface over theconterminous U.S. for each individual species.The grid used was the hexagonal grid of White etal. (1992), with some 12,600 points over the con-terminous U.S.

For the present project we have estimates ofland use changes for each county. We thereforeoverlaid the hexagonal grid on a county boundarylayer and determined the polygons generated byintersections of county and hexagon borders.Each polygon received the incidence value of its

3 One could argue that this does not represent the trueleast-cost solution since the program targets acres rather thancarbon (see Ref. Parks and Hardie, 1995). However, withineach state, there is little variation in carbon sequestration ratesacross counties, and the gains in efficiency from targetingcarbon would likely be outweighed by the additional costs ofadministering such a program.


source hexagon for that species and a county-wideincidence estimate was obtained by area-weightingthe incidence values for the polygons. To deter-mine how many birds would be lost from agricul-tural land or gained by new forests we consultedthe species lists of Lauber (1991), Peterjohn et al.(1993), Rodenhouse et al. (1995) to determinewhich species should be assigned to each of thesehabitats.4 Species not in either list were omitted,examples being shorebirds and wetland species.We then assembled the incidence data for all ofthe forest species and added the incidence valuesfor each county to get an index of abundance offorest birds since incidence measures are normallyproportional to absolute abundance (Hanski,1992). We repeated this for the species in the listof farmland species. Note that the resulting abun-dance measures for forest and farmland birdsassign equal weights to each species and, thus,assume that all species have the same conserva-tion value. Below, we discuss alternative ap-proaches that recognize differences inconservation importance.

In our calculations below we assume that birddensities remain constant, which is equivalent toassuming linear relationships between relativechanges in bird populations and percentagechanges in land use area. With rare exceptions(e.g. the house sparrow Passer domesticus) agri-cultural species do not display strong curvilinearrelationships with local habitat abundance(O’Connor et al., 1999). For many widespreadforest species, on the other hand, incidence fallsoff rapidly as forest stands break up (Askins,1993) and we, therefore, explicitly consider belowthe possible effects of this on our results.

To compute statewide estimates of bird popula-tion changes, we weighted each county’s results toaccount for the differential distribution of farm-land and forest and its birds across the state.Proportional changes in the habitat were multi-plied by the bird density in the habitat and thenby the area of habitat to arrive at a county changein bird population. To compute the proportional

change in birds at the state level, we summed thecounty changes in bird populations and dividedby the sum of the total populations of the coun-ties calculated in an analogous fashion.

3. Results

3.1. Land-use changes

Fig. 1 shows for each state the mid-1990s distri-bution of agricultural land and the distribution ofland that would convert to forestry under thestate-wide scenario of conversion of 10% of agri-cultural land for carbon sequestration.5 Currentland use and changes in land use are reported inpercentage terms to control for differences incounty land areas.6 In Wisconsin, agriculture isprominent in the southeastern parts of the state,particularly along a belt of counties betweenGreen Bay and Madison, and in a southern beltof counties bordering Illinois. However, the de-crease in agricultural land under the carbon se-questration policy is concentrated into thecounties with less intensive and profitable agricul-ture, being greatest in Jackson, Juneau, andAdams counties, and in a group of counties sur-rounding them in west central Wisconsin. Simi-larly, in South Carolina, agriculture isconcentrated in a broad band of counties acrossthe Atlantic Flatlands, with a scattering of moreproductive counties such as Anderson and Abbev-ille to the northwest and York in the north.However, the counties that would experience thegreatest relative loss of agricultural land—Georgetown and Berkeley on the Coastal Plainand Fairfield and McCormack inland—are out-side these areas, and most (though not all) of the

5 Plantinga et al. (1999) consider conversion rates rangingfrom 0 to 25% of state-wide agricultural land. The distributionof the estimated land-use changes across counties does notvary significantly with the state-level rate of agricultural landconversion.

6 For the same reason, we emphasize percentage changes inbird populations below. This makes our results easier tocompare across counties and states, though at the expense ofobscuring information on absolute changes.

4 Interested readers may contact the authors for a list ofscientific names and habitat classifications for the 159 speciesused in this analysis.


Fig. 1. Current distribution of agricultural land (left) and simulated change in agricultural land under a carbon sequestration policy(right) in southern Wisconsin, South Carolina, and Maine.

counties that would experience conversion rates of10–20% currently have less than 15% of their landin agriculture. Finally, in Maine, agriculture islargely concentrated in the southern coastal coun-ties, except for the potato lands of Aroostookcounty in the north, but with so little agriculture

in Maine most of these counties would also expe-rience conversion of land in pursuing the carbonsequestration policy. Land values in the two mostintensively agricultural counties—Androscogginand Kennebec—are such that conversion therewould be low, but all the other moderately farmed


Fig. 2. Current distribution of forests (left) and simulated change in forest area under a carbon sequestration policy (right) insouthern Wisconsin, South Carolina, and Maine.

counties except Waldo in the east would joininland Piscataquis and coastal Hancock countiesin disproportionate conversion.

Conversion of a given area of land to forestrycan have a small or a large relative effect on the

extent of forests in a county, depending on theexisting forest base. This is shown in Fig. 2 whichmaps both current forestlands and the relativeincreases brought about by the 10% conversionpolicy. The distribution of forests in Wisconsin is


Fig. 3. Frequency distributions of forest birds (left) and farmland birds (right) across counties in southern Wisconsin (a and b),South Carolina (c and d), and Maine (e and f).

largely complementary to that of agricultural landand concentrated in the northwest of the areamodeled here. As a result, the greatest relativeincreases in forest are not where the relative lossof agricultural land would be greatest (Fig. 1), butalong its southeastern fringe: the largest relativeincreases in forest lands occur in a belt of countiesextending northeastward and southwestward fromMadison. In South Carolina, on the other hand,very few counties have less than 50% of their landalready in forest, and the relative increases inforest would be only a few percent (i.e. an orderof magnitude smaller than in Wisconsin) andpatchy in distribution. Similarly, with so much ofMaine heavily forested the extra land in forestwould result in increases of only 1% or less formost counties (Fig. 2).

3.2. Bird distribution changes

The densities of birds within a county differedmarkedly between states, between habitats, andamong counties (Fig. 3). Median densities in Wis-consin were low in forests (median 12.2, range5.7–15.9), but much higher on agricultural land(median 31.5, range 27.5–33.3), but in South Car-olina were closer together (forest median density16.9, range 13.9–21.7; agricultural land density23.8, range 16.5–26.5). In Maine there was con-siderable overlap in densities in the two habitats(forest median 25.5, range 19.6–32.2; agriculturalland median 23.4, range 14.4–26.3). Examinationof Fig. 3 shows that forest densities were aboutequally variable in all three states, though withdifferent median densities, but agricultural bird


Fig. 4. Current distribution of farmland bird abundance (left) and projected change in farmland bird abundance under a carbonsequestration policy (right) in southern Wisconsin, South Carolina, and Maine. The metric I is Yang et al.’s (1995) index of totalbird abundance across multiple species (see text for details).

densities became more variable from Wisconsin toMaine, probably reflecting the increase in thevariability of agricultural conditions in the lessagricultural states.

Since bird densities are not uniform across eachstate, the consequences of land conversion forbird populations depend on the product of landuse changes and local bird densities. Althoughlocally the relati�e change in farmland bird num-bers must exactly mirror the relative changes in

agricultural land shown in Fig. 1, a 10% (say)change in the farmland bird population can in-volve a large absolute change or a small absolutechange, depending on the prevailing local densityof birds. Fig. 4 shows the spatial distribution offarmland birds within each state. In Wisconsin thedistribution of farm bird density largely resemblesthe distribution of agricultural land, being gener-ally high except within a cluster of eight countiesin the middle of the state. However, the two


distributions are not completely parallel and thearea of largest population decrease under thecarbon sequestration policy extends well beyondthese less intensively farmed counties (Fig. 4).Since these changes occurred over counties differ-ent in area and with different farmland bird densi-ties, a state-wide estimate of farmland bird lossrequired appropriate weighting of these effects,yielding a net reduction of 11.7% for Wisconsin(Table 1).

In South Carolina, farmland bird densities in-creased from the coast to the mountains ratherthan varying from county to county in directproportion to the intensity of agriculture in each(Fig. 4). The gradient was relatively shallow, how-ever, and as a result the changes in farmland birddistribution were largely determined by the rela-tive change in agricultural land: three of the fourcounties with greatest change in farmland birds—McCormack, Fairfield, and Berkeley—were alsoamong the top four for relative loss of agricul-tural land, and the patterns of change in agricul-tural land and in farmland bird distribution weregenerally similar (compare Figs. 1 and 4). Weight-ing these changes by area and size of current birdpopulation for each county gave a state-wide re-duction of 12.2% for South Carolina. In Maine,bird abundance generally decreases from south tonorth (Allen and O’Connor, 2000) and farmlandbirds, although largely paralleling the distributionof agricultural land, were correspondingly moreabundant in southern farming counties than in

northern ones (Fig. 4). As with South Carolina,the gradient was relatively shallow and thechanges in bird numbers largely reflected thechanges in agricultural land area. When weightedfor county area and distribution, the estimate ofthe state-wide decline was 10.8%.

Fig. 5 shows how the forest bird distributionwould change under a carbon sequestration pol-icy. In Wisconsin, the forest bird distributionlargely matches that of forests and the largestincreases are in counties with proportionatelylarge increases in forest land. However, withdense populations of forest birds across the north-west of the state, even modest increases in forestlands there result in large numerical increase (Fig.5). When this is coupled with the large amount ofagricultural land available for conversion underthe 10% scenario, the state-wide increase in forestbirds is very substantial, constituting a net in-crease of 22% (Table 1). In South Carolina, thedistribution of forest birds is strongly regionalwith highest densities in the Coastal Plain and inthe Piedmont, but as there would be relativelylittle increase in forest area in these parts of thestate under the carbon policy scenario, these areaswould contribute little to the state-wide popula-tion change. Instead, the largest change in forestbird abundance would be across the Atlantic Flat-lands, but as forest bird densities there are cur-rently rather low, the increases are also low. As aresult, when weighted for area and bird abun-dance, the state-wide change in forest bird popu-lations would be only 2.5% under the carbonsequestration policy. Similarly, in Maine, despitehigh densities of forest birds in the North Woods,the planting of additional forest within the extantfarming areas results in only minor increasesamong forest birds (Fig. 5), with a negligiblestate-wide increase of 0.3%.

3.3. O�erall bird population changes

Since the loss of agricultural land as habitatleads to a reduction in the farmland bird popula-tion in each county while the planting of newforest leads to a gain in forest birds, the netchange in bird numbers is a weighted function ofthe farmland and forest bird densities. Since the

Table 1Summary of percentage changes in forest and agricultural landarea and in population size of forest and farmland birds

Category Percentage changes

WisconsinMaine South Carolina

Forestland area 0.3 2.8 18.0−10.0Agricultural −10.0−10.0

land area3.2 2.5Forest birds 21.8

Farmland birds −11.7−10.8 −12.2−2.3Forest and −1.1−2.0

farmlandbirds


Fig. 5. Current distribution of forest bird abundance (left) and projected change in forest bird abundance under a carbonsequestration policy (right) in southern Wisconsin, South Carolina, and Maine. The metric I is Yang et al.’s (1995) index of totalbird abundance across multiple species (see text for details).

area changing in land use and farmland and forestbird densities all vary from county to county, thefigures have to be computed within counties andsummed to a state-wide total (Table 1). In fact, allthree states experience a net loss of birds: Wiscon-

sin loses 1.5%, South Carolina 2.3%, and Maine2.0%. That there should be a net loss even inWisconsin where the forest bird populations in-creased disproportionately largely reflects thehigher densities of farmland than of forest species.


Fig. 6. Frequency distributions of forest patch sizes for mixed coniferous–deciduous forests within cells of a 640 km2 hexagonal gridacross: (a) southern Wisconsin; (b) South Carolina; and (c) Maine.

3.4. Influence of forest patch size

In calculating the likely effects of afforestationon the populations of forest birds, it was assumedabove that new planting within any county wouldinduce only a pro rata increase in the local popu-lation of forest birds. Where existing forest isdistributed as a mosaic of small woodland blocks,however, new planting may coalesce these patchesinto larger stands of forest. Substantial evidence(Ambuel and Temple, 1983; Lynch and Whigham,1984; Askins, 1993; Hoover et al., 1995) exists toindicate that bird densities are often very muchlower in small patches of forest than in large ones(principally because predators and brood para-sites from the surrounding matrix can penetrate agreater proportion of small than of large patches).

Whilst much of the evidence derives from smallpatches some tens of hectares in size, we (R.J.O’Connor and L. Hayes in preparation) haveestimated the population losses associated withbreeding in forest stands of even some squarekilometers (rather than larger ones) to range from10 to 30% for several neotropical migrant birdspecies. In Fig. 6, therefore, we show the sizedistribution of the most common forest patchtypes in the three states, as derived from theremotely sensed data used by O’Connor et al.(1996). These data suggest that forest patch sizesin Maine and in South Carolina are generally solarge that patch size is unlikely to be a majorsource of further gain in forest bird numbers. InWisconsin, forest patches are much smaller andthe gains in forest bird populations estimated here


for Wisconsin are, therefore, likely to be minima,possibly to be increased by as much as 30% bycontiguous planting if the forest species there aregenerally area-sensitive. At present this is notknown.

4. Discussion and conclusions

To the extent possible, all relevant costs andbenefits should be considered in developing anational carbon sequestration strategy. Earliereconomic analyses of carbon sequestration pro-grams have focused on the opportunity costs ofagricultural production, but fail to account forpotential environmental effects of afforestation.In this study, we take a first step towards integrat-ing biodiversity impacts into the analysis. Theprincipal value of the results presented here is intheir quantification of factors readily identifiable apriori as potentially influencing the biodiversityconsequences of carbon sequestration policy.These included spatial variation in the density offorest and farmland birds, the relative extent ofagricultural land and forest cover within eachcounty, and the relative abundances of the farm-land species lost by land conversion to the forestspecies gained in the newly afforested habitat.

Just as one could anticipate on economicgrounds that the conversion of 10% of a state’sagricultural land to forest is unlikely to result in aconstant conversion rate across counties, onecould anticipate on biological grounds that birddensities were likely to differ across counties, in-troducing a spatial component to the pattern ofchange. The magnitude of this variation acrosscounties proved to be quite substantial (Fig. 3)and also to have marked spatial patterning (Figs.4 and 5). These results thus suggest that thebiodiversity consequences of afforestation as acarbon sequestration policy are unlikely ever to becaptured by simple pro-rating of constant densi-ties to the projected land changes. Moreover,particularly striking in Fig. 3 is the variation inrelative abundance of forest and farmland birdswithin the three states. The densities of forest andfarmland birds clearly respond differentially tovariation in environment between regions, and doso by quite significant amounts.

The relative abundance of forest and agricul-tural land cover within a county influences therelative impact of population changes amongforest birds. In each county a particular area ofland would convert from crops or pasture toforestry, and with a constant density of farmlandbirds within a county (assumed here) the relativechange in farmland bird populations is then nec-essarily pro rata. However, the relative effect onthe forest bird population in the county is thenweighted by the ratio of agricultural land to forestin the county. With twice as much agriculturalland as forest, a 10% loss of agricultural landyields a 20% increase in forest, with associatedincrease in the local forest bird population. Andthe converse applies.

Whether these land-use changes yield a netincrease or decrease in bird numbers also dependson the ratio of the local densities of farmland andof forest birds in the county. Even if a 10%change in agricultural land yielded a 20% increasein the area of forest, as in the example in theprevious paragraph, a net loss of birds resultsunless the density of forest birds is at least 75%that on agricultural land (since a 20% increase on75 yields 90, the reduced density of farmlandbirds). It is this effect that accounts for much ofthe net loss in bird numbers that our three studystates would experience under the carbon seques-tration policy modeled here. This is actually quitecounter-intuitive, in that common wisdom holdsthat loss and fragmentation of forest is a majorconservation issue for birds, and it is unexpectedto find fewer birds present after planting newforest! It appears that the density of birds onagricultural land may be higher in some statesthan is commonly acknowledged. Thus, both therelative densities of farmland and forest birds andthe relative extent of agricultural land and forestin the county determine whether the policy wouldresult in a gain or loss of avian abundance, withthe spatial variation between counties then deter-mining both the spatial patterning of the gainsand losses and, by virtue of the variation in areaamong the counties, whether the statewide out-come is a gain or a loss.

No consideration was given above to uncertain-ties in the measurement of the land cover propor-


tions and bird densities, but we expect the uncer-tainties in our estimates to be small. The baseland-use statistics are measured with very littleerror. The forest area measures are from U.S.Forest Service plot-level surveys, which have rela-tively small sampling errors. For example, thesampling error for timberland area in most Wis-consin counties (1996 inventory) is below 5%. Theagricultural land area measures are from the Cen-sus of Agriculture, which attempts to provide acomplete enumeration of the farm population.Response rates for the Census are very high and,thus, measurement errors are expected to be low.A second source of uncertainty arises with thepredictions of land use change. We do not expectthe prediction error to be large because of thegood fit of the econometric models and the factthat we consider modest changes in land use.

For bird densities the effects of uncertainty inincidence could be assessed by bootstrap samplingof the calculations, but a crude estimate suffices toshow that the effect is small. Uncertainty in inci-dence at a single location will be maximal for aspecies present in only half the surveys, whichwith our decadal estimate corresponds to a stan-dard deviation of 0.158 (= (0.5×0.5/10)1/2). Ifonly 40 species (about half of the farmland orforest bird species pools) were present in a loca-tion, the standard error for the farmland or forestbird incidence would then be only 0.025, negligi-ble as a fraction of any of the incidences reportedin Table 1 and an order of magnitude smallerthan the changes in overall populations discussed.In practice, the use by Yang et al. (1995) of datafrom multiple locations, the smaller standard de-viations of uncertainty in both common and rarespecies, and the generally larger species tallies ateach location mean that census uncertainties canbe neglected here.

We noted above that the predictions of forestbird changes might require modification to takeaccount of forest patch size distribution, particu-larly in Wisconsin. Askins (1993) found that pre-viously declining neotropical migrant speciesincreased in numbers as afforestation restored thecontiguity of forest in Connecticut. Hence, ourWisconsin estimates of the increase in forest birdpopulations must be seen as conservative: our

unpublished estimates indicate that populations ofneotropical migrants may be 30% lower in areasof forest fragments than where the forest is con-tiguous. If this were true of the Wisconsin forestspecies, forest populations could potentially in-crease by an additional 43% (=100/(100−30)) ifthe new forests were planted to maximize contigu-ity and reduce patch edge predation andparasitism.

A further assumption in our calculations wasthat there were no threshold effects in the influ-ence of forest density on birds. It is in principlepossible that there might need to be some mini-mum density of forest in an area before a forestspecies would settle there, and such an effectwould mean that newly forested land in countieswith little or no previous forest planting might notyield the expected gain in forest birds. However,the density estimates for forest species that weused were estimated from empirical bird distribu-tion data by Yang et al. (1995) using a grid ofsufficient resolution to average four points percounty. Accordingly, it is likely that any sucheffects present have already been incorporatedinto our analysis.

Our estimates of bird abundance were obtainedby summing estimates of incidence— the propor-tion of surveys at a site that recorded the spe-cies—over the 1981–90 decade. It is wellestablished that incidence is directly proportionalto absolute densities where estimates of both havebeen available, except for a small number of verycommon ubiquitous species whose densities mayvary within an incidence of 1.0 (O’Connor andShrubb, 1986; Hanski, 1997). If the constant ofproportionality were the same for all species, oursum of incidence values for forest species and forfarmland species would be directly proportionalto the sum of the corresponding bird densities,and if the value of the constant were known wecould re-express our measure as true densities. Itis, however, unlikely that all species share thesame constant and this leaves us with a possiblebias. Our sum of incidence measure would thenreally be of form � kiDi, where Di is the density ofthe ith species and ki is the constant of propor-tionality between incidence and density for thatspecies. Hence, were the constant for some partic-


ular species to be markedly higher than for theother species, then our summed incidence metricwould be higher for all points within the range ofthat species than it should be for proportionalityto true total density. However, our metric wassummed over many forest and farmland species,so any single error would be relatively small ineffect: with 50 species, even a doubling of theconstant of proportionality for one species wouldinduce an error of just two parts in a hundred. Inaddition, with so many species individual errorswere likely to cancel each other. We, therefore,consider the likely magnitude of any error fromthis source to be rather small relative to the effectsmeasured.

One important issue not considered explicitlyhere is that different species have different conser-vation significance. Our present calculationstreated all species as equally important, and com-puted the net change in birds under a carbonsequestration policy as the sum of the forest andfarmland bird changes in a county. In practice,some of the species lost from the newly forestedagricultural land will be of greater conservationvalue than the forest species gained. In principle,the analysis presented here could be conductedusing an index of conservation value instead ofthe incidence metric. For example, the incidencevalue for each species in a county could beweighted to reflect its relative conservation value,and the calculations that were here applied to justtwo groups— total forest birds and total farmlandbirds—could instead be computed over each ofthe entries in vectors of individual species inci-dence values for forest and for agricultural land.Applying the conservation weights to the resultswith and without the sequestration policy wouldthen yield estimates of the magnitude and distri-bution of conservation impacts. Several of theconservation value weighting schemes devised forbiodiversity complementarity analyses (Polaskyand Solow, 1995; Csuti et al., 1997) could readilybe adopted for use in the present context. As well,willingness-to-pay estimates from non-market val-uation studies could be applied to derive explicitestimates of the benefits or costs associated withchanges in bird populations.

As noted above, carbon sequestration is theobjective of the policy considered in this studyand the scenario we evaluate is designed toachieve a given level of land enrollment at theleast cost. There are several ways in which biodi-versity impacts—and other environmental effectsof afforestation—can be incorporated into theanalysis of carbon sequestration policy. If esti-mates of all the relevant benefits are available,including the benefits of climate change mitigationand biodiversity, then a standard cost-benefitanalysis can be performed. In this case, the opti-mal policy is to enroll land until the marginal costof enrollment equals the sum of marginalbenefits.7 If, as in the present case, reliablebenefits estimates are elusive because of the com-plexities of the natural phenomenon involved, analternative is to conduct cost-effectiveness analy-sis. This might involve minimizing the total costof enrolling land subject to constraints on theminimum amount of carbon sequestered and themaximum tolerable impacts on biodiversity. As-suming the biodiversity constraint binds, the solu-tion identifies the cost of departing from aleast-cost carbon sequestration strategy in orderto accommodate biodiversity objectives. Theshadow value on the constraint gives the implicitprice for biodiversity and, thus, sheds light on thenature of the tradeoffs involved.

Acknowledgements

The authors wish to thank our BiodiversityResearch Consortium collaborators B. Jacksonand S. Timmons (Oak Ridge National Labora-tory) and Carolyn Hunsaker (USDA Forest Ser-vice) for provision of landscape metrics; R.Neilsen, D. Marks, J. Chaney, C. Daly, and G.Koerper (USEPA Environmental Research Labo-ratory, Corvallis, OR) for assistance in computingclimate data; Tom Loveland (EROS Data Center,

7 For a given amount of land conversion, the marginal costof enrollment equals the corresponding per-acre subsidy that isused in the simulations. The marginal cost of enrollmentreflects the opportunity cost of agricultural production. SeeRef. Plantinga et al. (1999) for more details.


USGS) for land cover class data; and Denis White(Geosciences, Oregon State University) for assis-tance with spatial analysis. Financial support forthis research was provided through a CooperativeAgreement (PNW 00-CA-11261975-084) betweenthe University of Maine and the USDA ForestService (Andrew Plantinga, Principal Investigator)and between the University of Maine and OregonState University (Raymond O’Connor, PrincipalInvestigator).

References

Adams, D.M., Alig, R.J., McCarl, B.A., Callaway, J.M.,Winnett, S.M., 1999. Minimum cost strategies for seques-tering carbon in forests. Land Economics 75 (3), 360–374.

Adams, R., Adams, D., Callaway, J., Chang, C., McCarl, B.,1993. Sequestering carbon on agricultural land: social costand impacts on timber markets. Contemporary Policy Is-sues XI, 76–87.

Alig, R., Adams, D., McCarl, B., Callaway, J., Winnett, S.,1997. Assessing the effects of global change mitigationstrategies with an inter-temporal model of the U.S. forestand agricultural sectors. Environmental and Resource Eco-nomics 9, 259–274.

Allen, A.P., O’Connor, R.J., 2000. Interactive effects of landuse and other factors on regional bird distributions. Jour-nal of Biogeography 27 (4), 889–900.

Ambuel, B., Temple, S.A., 1983. Area dependent changes inthe bird communities and vegetation of southern Wiscon-sin USA forests. Ecology 64, 1057–1068.

Askins, R.A., 1993. Population trends in grassland, shrubland,and forest birds in eastern North America. Current Or-nithology 11, 1–34.

Birdsey, R.A., 1992. Carbon Storage and Accumulation inUnited States Forest Ecosystems. Washington, DC: U.S.Department of Agriculture, Forest Service, Gen. Tech.Rep. WO-59.

Csuti, B., Polasky, S., Williams, P.H., Pressey, R.L., Camm,J.D., Kershaw, M., Kiester, A.R., Downs, B., Hamilton,R., Huso, M., Sahr, K., 1997. A comparison of reserveselection algorithms using data on terrestrial vertebrates inOregon. Biological Conservation 80 (1), 83–97.

Hanski, I., 1992. Inferences from ecological incidence func-tions. American Naturalist 139, 57–62.

Hanski, I., 1997. Predictive and practical metapopulationmodels: the incidence function approach. In: Tilman, D.,Kareiva, P. (Eds.), Spatial Ecology: The Role of Space inPopulation Dynamics and Interspecific Interactions, pp.21–45 Monographs in Population Biology 30.

Hardie, I.W., Parks, P.J., 1997. Land use with heterogeneousquality: an application of an area base model. AmericanJournal of Agricultural Economics 77, 299–310.

Herkert, J.R., 1994. The effects of habitat fragmentation onmidwestern grassland bird communities. Ecological Appli-cations 4, 461–471.

Holdren, J., Lee, H., 1999. Workshop on Research and PolicyDirections for Carbon Management: Rapporteur’s Report.Harvard University, John F. Kennedy School ofGovernment.

Hoover, J.P., Brittingham, M.C., Goodrich, L.J., 1995. Effectsof forest patch size on nesting success of wood thrushes.Auk 112, 146–155.

Lauber, T.B., 1991. Birds and the conservation reserve pro-gram: a retrospective study. M.S. thesis, University ofMaine, Orono, ME.

Lichtenberg, E., 1989. Land quality, irrigation development,and cropping patterns in the northern high plains. Ameri-can Journal of Agricultural Economics 71, 187–194.

Lynch, J.F., Whigham, D.F., 1984. Effects of forest fragmen-tation on breeding bird communities in Maryland, USA.Biol. Conserv. 28, 287–324.

Moulton, R., Richards, K., 1990. Costs of Sequestering Car-bon through Tree Planting and Forest Management in theU.S. Washington, DC: U.S. Department of Agriculture,Forest Service, Gen. Tech. Rep. WO-58.

National Academy of Sciences. 1992. Policy Implications ofGreenhouse Warming: Mitigation, Adaptation, and theScience Base, National Academy Press, Washington, DC.

Newell, R.J., Stavins, R.N., 2000. Climate change and forestsinks: factors affecting the costs of carbon sequestration.Journal of Environmental Economics and Management 40,211–235.

O’Connor, R.J., Boone, R.B., Jones, M.T., Lauber, T.B.,1999. Linking continental climate and land use patternswith grassland bird distribution in the conterminousUnited States. Studies in Avian Biology 19, 45–59.

O’Connor, R.J., Dunn, E., Johnson, D.H., Jones, S.L., Petit,D., Pollock, K., Smith, C. R, Trapp, J.L., Welling, E.,2000. A Programmatic Review of the North AmericanBreeding Bird Survey: Report of a Peer Review Panel.Report to Patuxent Wildlife Research Center, Laurel, MD.(http://www.pwrc.usgs.gov/bbs/bbsreview/bbsfinal.pdf).

O’Connor, R.J., Jones, M.T., White, D., Hunsaker, C., Love-land, T., Jones, B., Preston, E., 1996. Spatial partitioningof the environmental correlates of avian biodiversity in thelower United States. Biodiversity Letters 3, 97–110.

O’Connor, R.J., Shrubb, M., 1986. Farming and Birds. Cam-bridge University Press, Cambridge.

Parks, P.J., Hardie, I.W., 1995. Least-cost forest carbon re-serves: cost-effective subsidies to convert marginal agricul-tural land to forests. Land Economics 71, 122–136.

Peterjohn, Bruce G., Sauer, John R., 1993. North Americanbreeding bird survey annual summary 1990–1991. BirdPopulations 1, 52–67.

Plantinga, A.J., Mauldin, T., 2001. A method for estimatingthe costs of CO2 mitigation through afforestation. ClimaticChange 49, 21–40.

Plantinga, A.J., Mauldin, T., Miller, D.J., 1999. An economet-ric analysis of the costs of sequestering carbon in forests.American Journal of Agricultural Economics 81, 812–824.


Polasky, S., Solow, A.R., 1995. On the value of a collection ofspecies. Journal of Environmental Economics and Man-agement 29 (2), 298–303.

Richards, K.R., Moulton, R., Birdsey, R.A., 1993. Costs ofcreating carbon sinks in the U.S. Energy Conservation andManagement 34, 905–912.

Robbins, C.S., Dawson, D.K., Dowell, B.A., 1989a. Habitatarea requirements of breeding forest birds of the MiddleAtlantic states. Wildl. Monogr. 103, 1–34.

Robbins, C.S., Sauer, J.R., Greenberg, R.S., Droege, S.,1989b. Population declines in North American birds thatmigrate to the neotropics. Proceedings of the NationalAcademy of Sciences of the United States of America 86,7658–7662.

Robinson, S.K., Thompson, F.R. III, Donovan, T.M., White-head, D.R., Faaborg, J., 1995. Regional forest fragmenta-tion and the nesting success of migratory birds. Science 67,1987–1990.

Rodenhouse, N.L., Best, L.B., O’Connor, R.J., Bollinger,E.K., 1995. Effects of agricultural practices and farmland

structure. In: Martin, T.E., Finch, D.M. (Eds.), Status andManagement of Neotropical Migratory Birds. Oxford Uni-versity Press, New York, pp. 269–293.

Stavins, R., 1999. The costs of carbon sequestration: a re-vealed-preference approach. American Economic Review89, 994–1009.

Vickery, P.D., Hunter, M.L. Jr., Melvin, S.M., 1994. Effects ofhabitat area on the distribution of grassland birds inMaine. Conservation Biology 8, 1087–1097.

White, D., Kimmerling, A.J., Overton, W.S., 1992. Carto-graphic and geometric components of a global samplingdesign for environmental monitoring. Cartography andGeographic Information Systems 19, 5–22.

Wu, J., Segerson, K., 1995. The impact of policies and landcharacteristics on potential groundwater pollution in Wis-consin. American Journal of Agricultural Economics 77,1033–1047.

Yang, K.-S., Carr, D.B., O’Connor, R.J., 1995. Smoothing ofbreeding bird survey data to produce national biodiversityestimates. Computing Science and Statistics 27, 405–409.

quantifying the impacts on biodiversity of policies for carbon sequestration in forests

Documents